GeForce GTX 480 PCI-Express Scaling

GeForce GTX 480 PCI-Express Scaling

(25 User comments) »

Conclusion

From time to time, being able see the impact of running high-end graphics cards with lesser system bus bandwidth is important for many reasons. From an academic standpoint it lets you see how far current day GPUs have gone about making the most out of abundant interconnect bandwidth. It also is testament to the validity of motherboard maker arguments who want to sell you high-end motherboards with features that boost bandwidth using mumbo jumbo terms such as PCI-Express switches with data broadcast features. On Intel's LGA 1156 and most AMD motherboards there is one PCI-Express x16 link that ends up being split in two x8 links for each of the two graphics cards. Our testing offers you data to decide whether opting for the more expensive motherboard that uses a chipset with full-bandwidth x16 links for each of the graphics cards is really worth it; or also shelling out more on motherboards with fancy ways to step up bandwidth (using PCI-E switch chips). Our PCI-Express 2.0 x8 results can also be used to judge PCI-Express 1.1 x16 performance (which offers the exact same bandwidth). Many motherboards using older chipsets such as Intel P35, NVIDIA nForce 680 SLI, etc, have only support for version 1.1 of the PCI-Express interface.

That it did. NVIDIA's GeForce GTX 480 is a very fast graphics processor. To maintain its speeds, it would hypothetically require high system bandwidth, leading one to think that lesser PCI-Express configurations would cripple it. The theory couldn't be more wrong, as seen by the mere 2% performance loss going from x16 to x8 (which reduces bandwidth by 50%). To cite results from one of the latest and resource-heavy games in our bench, Collin McRae DiRT 2, that translates into something like 63.2 FPS vs. 62.1 FPS, at 2560 x 1600 pixels resolution - barely a difference.
We also examined how much PCI-E 2.0 x4 (or PCI-E 1.1 x8, older motherboards) would affect the GTX 480. And the result coarsely put is "not much": 8%. The other part of the analysis is to see how the GTX 480 scales with decreases in interconnect bandwidth compared to AMD's Radeon HD 5870. When compared to the card from the red camp, the GeForce GTX 480 takes bigger hits with decrease in bandwidth. It is scaling worse than the HD 5870, which could be because the GPU is faster and makes better use of available bandwidth, and hence gets the pinch due to lack of it. That still shouldn't come in the way of you using two GTX 480 cards in SLI on a motherboard with electrical x8 PCI-E slots. You shouldn't be worried about using these cards on a PCI-E 1.1 motherboard, either.
Page:
Next Page »(25 User comments)